In 1995, Football Outsiders graded the Eagles special teams as the worst in the NFL. The next two years, Philadelphia ranked 20th and 26th, respectively. In 1998, after hiring a new special teams coordinator, the team still finished just 25th. But, over the next eight years, the Eagles’ special teams flipped dramatically, ranking as the second-best in football during that period. In fact, from 2000-2004, Philadelphia ranked in the top five in the Football Outsiders’ special teams ratings each season.

When the Ravens hired the coordinator of those special teams, John Harbaugh, as their head coach in 2008, Baltimore turned one of the more surprising coaching hires in recent history into one of the best. Based on where the team was when it hired him, Harbaugh’s first three years were about the best since 1990 of any coach not named Harbaugh, at least according to DVOA. The Ravens made the playoffs in Harbaugh’s first five seasons, winning the Super Bowl in the last of those. Harbaugh’s success even caused Chase to wonder whether it would change the way teams hired head coaches.

Since Harbaugh was so successful as a coordinator, does that mean he was a good bet to be a successful head coach? At first glance, you might think just about every coordinator who gets promoted or poached to become a head coach was very successful in his previous job. As it turns out, that’s not always the case. Once we correct for expectations, a little more than one in four hired head coaches actually underperformed in their previous jobs, at least according to DVOA.

Consider one man who performed particularly poorly as a coordinator: Eric Mangini. The 2005 New England defense had a DVOA that was 15.2 points lower than we would have predicted based on the Patriots’ performance in the preceding seasons. He was not so much of a (Man)genius to have a good defense in 2005, and that may have given some hint that he was not the greatest bet to succeed as a head coach, either. ^[1]Always a bonus when painful Jets memories come up organically. There are always other coaching greats like Joe Walton for Jets fans to remember fondly, at least for epic nasal invasions.

This leads to an obvious question: on average, have teams done better when they have hired head coaches who were actually good in their previous jobs (either as coordinators or head coaches)? Let’s take this to the data.

The Relationship Between Prior Performance and Future Performance for Coaches

The table below describes the idea behind my projections of future success based on prior performance. It has Mangini’s actual and predicted performance in his three jobs as a head coach or coordinator. ^[2]All of the data on coaching tenure come from Pro Football Reference and the data on DVOA from Football Outsiders, of course. Since I want positive numbers to indicate better performance for offense, special teams, and defense, I flipped the sign on all the defensive DVOA ratings. ^[3]The default, of course, is that negative DVOA ratings are better for defenses.

The first line refers to Mangini’s one season as Patriots defensive coordinator. The Pats had a bad year, ranking 27th and with a DVOA of 9.2% below average. That rating looks even worse in light of their good performance in previous seasons. Based on those previous years, I come up with an expected DVOA of 6.0 for the Patriots. ^[4]I derived this estimate by running a regression of DVOA in the current year on DVOA in the three previous ones. I ran separate regression for overall DVOA, offense, defense, and special teams. So I assigned Mangini’s one year in charge of the Patriots’ defense a Value Above Expectation (VAE) of -15.2. Even though his Browns team had a worse average overall DVOA in 2009-2010, his VAE is not as bad because those Browns projected poorly based on previous performance.

I derived a VAE for each coaching and coordinator tenure since 1992. To create apples-to-apples comparisons for coaches who had varying durations of coaching tenures, I looked only at a coach’s first three years in a job to come up with his VAE. ^[5]There’s no perfect decision in this regard, however. Gus Bradley, for example, gets underrated as a defensive coordinator because 2012 was his fourth year in that job with the Seahawks. To predict future success, I then performed a regression to predict a head coach’s VAE based on performance in his three previous jobs. To measure previous performance, I weighted a coach’s previous VAE by time served. For example, look at Mike McCarthy’s coaching jobs. His one year as the 49ers offensive coordinator was a disaster. The Niners finished dead last in offensive DVOA and McCarthy’s VAE for that year was -33.3. But McCarthy’s was much better in his first three years as Saints’ offensive coordinator relative to expectation. Since that stint gets three times more weight than his single year in San Francisco, McCarthy’s previous performance measure is, while not good, actually not terrible. His average previous performance as a coordinator before he took the Packers’ job works out to -3.3.

Across all head coaches, previous performance does seem to project future success, at least a little bit. An additional one point of previous success predicts that a head coach does about 0.2 points better. In other words, head coaches on average carry about 20% of their previous success into their current jobs.

But I noticed something interesting when I delved a little deeper into the data. It turns out that all of this is driven by the people in the sample who were previously head coaches. The first graph below looks just at first time head coaches, which means only data on earlier stints is limited to time as coordinators. The second one looks at people who were previously head coaches. In the first graph, the data points are just the head coaches who started their jobs since 2004 to keep it easy to read and the line refers to the sample of all coaches. Each data point is identified by the head coach’s first initial, first two letters of his last name, and the year the head coaching job started. BPa-03, for example, refers to Bill Parcells’ job with Dallas that started in 2003.

The first figure shows that success as a coordinator does not seem to tell us very much about success for first-time head coaches. On the other hand, the relationship between previous performance and future success is strong for previous head coaches. For second (or third or fourth) acts, head coaches carry over about 70% of their value from their previous jobs. ^[6]The regression coefficient is 0.70 and the p-value is .011, where I correct the p-value for any coaches who appear more than once.

The simplest reading of this result―and one I think is at least part of what’s going on―is that coaches who were produced good performance in the past continue to perform well in the future. Coaches who failed before continue to fail. ^[7]Note that Bill Belichick’s tenure in Cleveland came before the start of this study. Out of all second-run coaches who had negative performances according to VAE in their previous jobs, the only coaches that generated positive VAE in their next jobs were Marty Schottenheimer and Ted Marchibroda. Schottenheimer, in particular, is an exception who just shows the limitation of the time frame of the data. He gets no credit here for his tenures with the Browns and Chiefs because his tenures started before 1992. So his negative previous performance is just based on his one year in Washington. Since 1992, there really is no example of a coach who failed overall in his previous jobs and then succeeded as a second-run head coach. ^[8]Note that by my definition, Pete Carroll was not a failure in New England. When I’m the GM of the next Los Angeles NFL team, you can assume I will not be hiring Mangini as my head coach.

The table below has the VAE estimates and previous performance measures for the full set of head coaches.

Predicting Coaching Success and Longevity

I am working on a coaching projection system, one piece of which is the idea described here. ^[9]If I am allowed to create acronyms, my current thinking is CEFALO = Coaching Efficiency Forecast And Longevity O? I’m not sure what to do with the O. Organization? Oligopoly? Owl? Teams that hire retread coaches who have failed in their previous head coaching jobs end up with coaches who fail for them. Why do they hire those coaches in the first place? One possibility is that those coaches got lucky in their previous jobs, something that DVOA catches but teams may not. Another possibility is that coaches benefit from regression to the mean, getting credit for improvements that they did not cause.

In addition, the best coaches may have more of a chance to pick their spots. Tony Dungy is a good coach — hey, he got an Index named after him for a reason — but his big VAE in Indianapolis happened in part because he went to a good situation that was likely to improve for any coach. He also had a good general manager, at least at that time. ^[10]Bill Parcells always had a knack for putting himself in a spot where there was nowhere to go but up. On the other hand, coaches such as Chan Gailey had much less power to get themselves into a good situation. Still, it seems revealing that head coaches who have failed before almost never work out. The data suggest that the Bills made a better move going with say, Doug Marrone, over a known non-success like Gailey, since that strategy basically never works.

If we are looking at the recent coaching hires that project the best according to previous performance, we would be bullish on Lovie Smith, meh on Ken Whisenhunt, and bearish on Jim Caldwell. Smith succeeded both as Rams defensive coordinator and Bears coach, while Caldwell had a negative VAE both as Colts head coach and Ravens offensive coordinator. ^[11]Of course, a pretty important part of his resume is not included in his VAE calculation. Of course, this is just one piece of projecting coaching success and more on the other pieces coming soon.